Advanced and labor-saving measures required for package appearance inspection.

In recent years, there has been a strong demand for advanced quality control and labor-saving measures in manufacturing facilities for food, pharmaceuticals, and daily necessities.

In particular, how quickly we can detect issues such as poorly printed product labels, mix-ups of different products, and damaged packaging is a crucial issue in protecting product quality and brand value.

Traditionally, many manufacturing lines have relied on human visual inspection and image inspection.

However, in recent years, the diversification of product lineups and the shortening of product cycles have resulted in a significant amount of work being required to adjust and readjust inspection conditions.

In this context, one approach that is attracting attention is the use of AI, specifically OCR (Optical Character Recognition) and image recognition, to automatically recognize characters and codes from camera images for visual inspection. 　

In fact, the market for AI-powered package visual inspection machines is projected to expand from $1.6 billion in 2025 to $3.2 billion in 2035, representing approximately double the growth over the next 10 years. (*)

*Source: Future Market Insights Inc, “AI Powered Packaging Inspection Machine Market Size and Share Forecast Outlook 2025 to 2035”

This article introduces a case study of next-generation edge AI applications using SiMa.ai, based on a demonstration of AI-powered visual inspection utilizing OCR and image recognition for food packaging.

Case study: Automation of package inspection using OCR (optical character recognition) and image recognition.

In manufacturing lines for food, pharmaceuticals, and other products, checking the printed content, product type, and label condition of the packaging labels is a routine procedure.

In particular, incorrect packaging, mixing of different product types, and damaged labels are significant issues that can lead to shipping errors and recall risks.

In this demo, an AI running on an edge device performs OCR (optical character recognition) and image recognition on camera footage in real time to determine the product's condition on the spot.

For example, for yogurt products flowing through a production line,

Is the product label correct?

- Check if the label is damaged.

- Check if different varieties are mixed in.

We will inspect it in real time.

Normal products are displayed in green, and defective products in red, allowing for intuitive identification of abnormalities on the production line.

Furthermore, by immediately detecting label damage or the inclusion of different product types, it contributes to the early reduction of quality problems and compliance risks.

Here is a demonstration of package inspection using OCR and image recognition (video in English).

In this demo, the following processes are performed on the edge device:

① Extraction of label text information using OCR

② Product logo/image recognition using image recognition

③ Real-time detection of mismatched products and damaged labels

④ Real-time tracking of products moving along the production line

⑤ Complete all processing within the local environment.

This allows for real-time decision-making on the spot, without relying on a cloud connection.

Furthermore, in addition to stable operation that is less affected by communication environments, it also offers security advantages because there is no need to transmit confidential information externally.

Flexible AI visual inspection achieved through a template-based approach

One of the key features of this system is its ability to flexibly configure inspection settings using a template-based approach.

In conventional AI-based visual inspection systems, it was often necessary to perform additional training and model adjustments every time a new product or label change was introduced.

On the other hand, in this demo, you can specify a region of interest (ROI) on the package and perform the inspection by setting the text or images you want to check as a template.

Therefore, there is no need to develop a new AI model for each application, and settings can be changed in a short amount of time.

Furthermore, this demonstration uses only a commercially available USB camera and ambient lighting, and does not require any special lighting.

This system has a configuration that makes it easier to reduce the equipment load and on-site adjustment man-hours during implementation compared to conventional visual inspection systems.

This solution has the following features:

• A configuration that does not require expensive GPU servers or dedicated lighting.
• It can be implemented using commercially available USB cameras.
• It is possible to change the inspection settings without performing a large-scale retraining.
• Integrating OCR and image recognition on a single SoC
• Low power consumption operation of 10W or less
• High-speed line response with real-time judgment

Summary: AI visual inspection is becoming a "system usable in the field."

In the case study presented here, combining OCR and image recognition enables real-time detection of issues such as mismatched products and damaged labels, thereby automating quality control on the manufacturing line.

Furthermore, because inspection settings can be flexibly configured using templates, configuration changes can be made quickly even when the product changes.

Furthermore, the simple configuration using a single SoC enables low-power and easy-to-implement AI-powered visual inspection.

Edge AI-powered visual inspection is not merely a tool for improving operational efficiency; it is expected to be used even more extensively in the future as a system that contributes to advanced quality control and reduces the burden on on-site personnel.

​The SiMa.ai​ ​MLSoC™ used in this demonstration is a next-generation chip optimized for realizing AI that can be used in the field.

Despite its compact size and energy-saving design, it achieves highly efficient inference of up to 50 TOPS and supports high-speed production lines with real-time processing at 120 FPS. The flexible development environment provided by the built-in Arm Cortex-A65 is also a major benefit of its introduction.

Those considering introducing AI can start small by beginning with an inspection process like the one in this example, which will allow them to verify its applicability to their own production lines and the potential for improvement.

We hope you will find this article useful as your first step in utilizing AI.

Inquiry

Please feel free to contact us with any questions about our products, technical inquiries, sample requests, or estimates.